fix: fail when diloco is used with global pg size of 1

src/zeroband/diloco.py

@@ -58,6 +58,8 @@ def __init__(
         self._init_offloaded_optimizer(model=model)
 
     def _init_offloaded_optimizer(self, model):
+        if self.global_pg.size() < 2:
+            raise ValueError(f"Diloco requires at least two nodes to work. Got a world info of {self.world_info}")
         self.param_list_cpu = self.get_offloaded_param(model)
         self.outer_optimizer = torch.optim.SGD(
             self.param_list_cpu, lr=self.config.outer_lr, momentum=0.9, nesterov=True